What is the number of GFlops per cpu cycle for IBM Power8?
Flops per cpu cycle is obtained as ( number of double precision floating point numbers fitting into one vector processing unit (if any) ) * (# Vector processing units) * 2 (if FMA is supported).
What is it for a Power8 processor? I read that Power* currently has one vector processing unit per core, and it is supporting 4 double precision numbers in it. so for instance for Power S822LC model 8355-GTA I find a peak performance of (8 cores) * (3.32 GHz) * (4*2) = 212 GFlops per socket. Could someone confirm? In my calculation, 4*2 is the number of Flops per cpu cycle, 4 is the number of double precision number in register and the factor 2 is for FMA support.